Tools that track AI coding spend across a team: TokenShift, Vantage, CloudZero, Faros AI, Jellyfish, Weave, Requesty, Torii and OpenReason Intel

TokenShift, by PointFive

PointFive calls TokenShift "the AI agent control plane for your employee workstations". It "runs natively on employee workstations, at each AI agent, deployed through the MDM you already use", naming Jamf, Intune and JumpCloud, and says "Data is processed on the device. Prompts never leave the machine for a central store." It breaks "cost, active users, sessions and tokens down by AI agent, team and model", tied to person, repo, pull request and ticket. It goes further than seeing: teams "Set budgets by organization, team, user or model, and choose how TokenShift responds when a limit is reached: warn, block, or require approval", and can allow or block models and tools. It suits companies whose IT team already manages laptops centrally. Pricing is on request.

OpenReason Intel

OpenReason Intel is a dashboard for Claude Code and Codex usage. Each engineer runs one command from the or-otel quickstart on a Mac, Windows or Linux machine, which installs a reporter; requests still go straight to Anthropic and OpenAI. The dashboard shows each person's 5-hour and weekly window history, the periods when a window ran out with the session, repo and model around them, seat commitments beside variable spending, and branch activity beside recorded pull requests. Its limits: it covers Claude Code and Codex only, with OpenCode support coming; it sets no budgets and does not notify or stop anything, though a team with a coding agent can build its own warning on the data; and its costs are reported, allocated or estimated rather than an invoice. It is free under 10 engineers and $20 an engineer a month above that (pricing).

Vantage

Vantage is a cloud cost platform. Its Cursor integration promises to "break down spend by developer and model", connected with an API key, and its guide says "You'll need a Cursor Enterprise plan (the Admin API isn't available on lower tiers)", with "No agents to install". It also reads the Claude API and Claude Enterprise analytics, and Codex "by client, model, token type, and user". Its Anthropic and OpenAI pages offer cost or usage budgets that alert you when spending runs over. It suits finance and platform teams who already report on cloud costs. It sees what the vendors' APIs expose and nothing from engineers' machines. A free tier covers up to $2,500 of cloud spend, and paid plans start at $30 a month (pricing).

CloudZero

CloudZero "connects to AI platforms to bring your AI costs into a unified view alongside your cloud and SaaS spending", and "All connections are read-only." It reads Anthropic's platform and Enterprise analytics, OpenAI through an Admin API key, and Cursor through its Admin API; the overview says Cursor needs an Enterprise plan, while its Cursor page says Teams works too. It also reads Claude Code's own telemetry, and offers a macOS Collector, a menu-bar app that IT can roll out to every Mac, which can add branch and pull request fields. Its budgets send an alert when spending crosses a threshold. It suits finance teams that allocate AI spend the way they allocate cloud spend. Pricing is a custom quote.

Faros AI

Faros AI's guide to tracking AI coding costs across teams (26 August 2026) says it connects "Copilot, Claude Code, Cursor, Codex, and homegrown agents" and links AI coding sessions to "commits, pull requests, CI results, and production deployments". Its home page calls it a "Token Engineering platform" that "enforces your approved model routes, budget controls, and compliance policies through a model router". Faros says deployment "does not require any software installation on developer machines", because it reads the tools' built-in telemetry, and it lets each team track spend against a budget. It suits engineering leaders who need to show whether AI spending produced shipped work, and its pages stress teams rather than individuals. Pricing is based on consumption and is on request.

Jellyfish

Jellyfish AI Impact "measures how AI tools, like Copilot, Cursor, Claude Code, and agentic systems, actually affect software delivery". It shows spend and token usage by tool, team or initiative, and connects AI usage and spend to "throughput, cycle time, code quality". It names GitHub Copilot, Cursor, Claude Code, Amazon Q, Gemini Code Assist, Windsurf, CodeRabbit, Devin, Copilot Agent and Google Jules, and it "derives AI signals directly from Git, planning systems, and workflow data". Its page does not say where the spend figures come from, and it does not describe caps. It suits engineering executives and finance teams at companies that already run an engineering intelligence platform. Pricing is "based on number of seats and specific modules selected".

Weave

Weave, at weaveos.com, says "Weave builds models to measure engineering work". Its token intelligence product "tracks per-user costs for Claude Code, Cursor, GitHub Copilot, Codex, Devin, Windsurf, Amazon Q, Sourcegraph, Augment, Kiro, and Firebender", and labels which costs come from provider billing, which are modelled and which are estimates. For Cursor it uses an admin API key, on Teams or Enterprise, plus a Weave plugin that each engineer sets up for session telemetry. Its Pro plan adds a drill-down to pull requests. The pages read for this guide do not describe caps or blocking. It suits engineering organisations that want AI spend beside a measure of output. Plans start at $0, and Pro costs $50 an engineer a month.

Requesty

Requesty describes itself as "a gateway for the configured API traffic". Teams give each engineer their own Requesty key for API-billed Claude Code and Codex, and admins "set an individual monthly limit for each engineer" that members cannot change. Its guide suggests "80% and 95% as warning points" and says "Alerts notify an owner; limits control spending." It sees only the traffic routed through it, and says so: "Native subscriptions, Cursor Tab and Codex cloud still need separate accounting." It suits teams willing to pay for models per token through a gateway rather than through each vendor's subscription. You pay "what your application spends, plus 5%", and can start free.

Torii

Torii says it "discovers your entire SaaS and AI stack, reclaims wasted spend, and automates onboarding and offboarding", using finance, identity, HR and browser data. Its article on tracking Cursor says its AI dashboard slices Cursor usage by employee, model and time, names Claude Code and ChatGPT beside it, marks seats idle for 90 days or more for reclamation, and sends renewal alerts. The same article says Vantage is the only tool on its list that plugs straight into Cursor's Admin API, so how Torii gets per-model token figures is not clear from its pages. It suits IT and finance teams who mostly want to find every seat and every invoice. Pricing is not published.

The local trackers, ccusage, UsageAtlas, CodeBurn and OpenUsage, also come up in these answers. They are free, run on one person's machine and show that person's usage, so they do not answer the team question; the guide comparing the local trackers with OpenReason Intel covers them. CodeBurn has a teams page marked "Coming soon".

Pylon appears in some AI answers about team AI spend, but it is a customer-support platform. Its only related page answers whether using Claude, Devin or Cursor from inside Pylon counts toward those tools' own spend: "usage may still count toward that tool's own usage or spend". It does not track AI coding spend, so it is not described above.

• If you want to set budgets that stop spending on every laptop, and IT already manages laptops centrally, TokenShift says it does that for six coding agents.
• If you already pay for Vantage or CloudZero, you can add Cursor, Claude and Codex spend to the reports you have, with the plan requirements each one states.
• If you already run Faros AI, Jellyfish or Weave, you can see AI spend beside delivery measures in the platform you have.
• If you are willing to route model traffic through a gateway, Requesty gives per-engineer limits for the API-billed tools.
• If you mostly want to find seats and invoices, Torii is built for that.
• If your team uses Claude Code and Codex and wants each person's window history, usage past the plan, seats and recorded work in one place, OpenReason Intel does that, free under 10 engineers, without budgets or notifications.